The U.S. government's sudden decision to restrict access to Anthropic's Mythos and Fable large language models has become a watershed moment for India's sovereign AI ambitions, validating warnings from national security officials who have long argued that over-reliance on foreign AI technology poses strategic risks. On Friday evening, the U.S. ordered Anthropic to disable access to these models for non-U.S. nationals, even within the company itself, citing concerns about potential "jailbreaks" of the models' safety protections.

The timing is particularly significant for India, which had been seeking access to Mythos specifically for its cybersecurity capabilities. Mythos is a model that Anthropic claims excels at finding and patching software vulnerabilities that human researchers have missed for decades. India's Ministry of Electronics and Information Technology and other government agencies had joined a program called Project Glasswing earlier this month to gain access to these tools, but that access is now disrupted.

Why Does This Matter for India's AI Independence?

The export controls have triggered an "I told you so" moment among India's national security hawks. Sridhar Vembu, founder of Zoho and a member of India's National Security Advisory Board, declared on social media that "globalisation is dead" and urged India to embrace smaller, open-source AI models from both Indian and Chinese developers rather than depend on U.S. companies that may suddenly cut off access.

"Globalisation is dead, and Bharat must find its own way ahead," Vembu stated, adding that India should "ensure that organisations in India embrace smaller models, both Indian and Chinese open-source ones."

Sridhar Vembu, Founder of Zoho and Member of India's National Security Advisory Board

However, the path to true AI sovereignty is far more complicated than simply switching to alternative models. India's capacity to train frontier-class AI models lags significantly behind both the United States and China. While China has companies like DeepSeek that use large quantities of graphics processing units (GPUs) with abundant data centre capacity and electricity access, India faces severe resource constraints.

What Are the Real Barriers to India's Sovereign AI Development?

Creating a homegrown alternative to Mythos would require access to expensive AI chips from companies like Nvidia, massive data centre capacity, and reliable electricity supply. The costs run into the tens of billions of dollars, making the ambition difficult even for a nation of India's economic scale. Even Vembu, despite his calls for sovereignty, acknowledged the harsh reality of the situation.

"We must deepen our R&D, but remember that the latest models cost not only huge GPU budgets to train; the GPUs themselves are restricted. So we can't afford the scale of money (of the order of $100+ billion to even get in the game!), and even if we could come up with the money, we can't get all the GPUs," Vembu explained.

Sridhar Vembu, Founder of Zoho

Despite these constraints, some progress has been made. Bengaluru-based Sarvam AI launched a 105 billion parameter large language model (LLM) specifically trained with an Indian bias to counter the U.S.-centric nature of most AI models. However, this model remains far from frontier-class capability for ambitious cybersecurity work.

Steps to Accelerate India's AI Sovereignty

• Increase Government Funding: T.V. Mohandas Pai, a former CFO of Infosys and government technology consultant, called for an annual fund of 50,000 crore rupees (roughly $6 billion USD) dedicated to deep tech and AI research, plus a 200,000 crore rupee emergency credit guarantee fund to build hyperscale cloud infrastructure, hardware, and chips.
• Embrace Smaller, Open-Source Models: Rather than attempting to build frontier models from scratch, India could focus on optimizing and deploying smaller, open-source models that are less resource-intensive but still capable of handling critical tasks like cybersecurity and government applications.
• Invest in Lower-Cost Research: Vembu advocated for research initiatives like those undertaken by Zoho, which recently announced an indigenously developed server, as a more pragmatic path than attempting to match U.S. spending on frontier models.

The disruption extends beyond government agencies. Vikram Chandra, an entrepreneur and journalist, noted on social media that his projects relying on Fable, which had been available to paying users of Anthropic's Claude AI assistant, would "come to a grinding halt" due to the restrictions. He acknowledged that guardrails for frontier AI are essential but argued that "creating national barriers isn't the solution".

The broader context reveals a global shift toward sovereign AI development. According to market research, the global sovereign AI market was valued at approximately $40 billion in 2025 and is projected to reach $148 billion by 2032, growing at a compound annual growth rate of 20.6 percent. This expansion reflects governments worldwide investing billions in domestic AI capabilities, including national data centres, AI supercomputing infrastructure, sovereign cloud environments, and locally governed AI models.

Countries including Canada, France, Singapore, the United Kingdom, and India are all pursuing sovereign AI initiatives, with these projects requiring substantial investments in GPUs, networking systems, storage infrastructure, and AI development tools. The shift signals that nations increasingly view AI infrastructure as strategic national infrastructure, much like transportation networks or energy systems.

For India specifically, the U.S. export controls serve as a stark reminder that technological independence cannot be taken for granted. While the immediate impact disrupts access to cutting-edge cybersecurity tools, the longer-term challenge is building the institutional capacity, funding mechanisms, and technical expertise to develop sovereign AI capabilities that can compete on the global stage. The question now is whether India's government and private sector can mobilize resources quickly enough to avoid becoming permanently dependent on foreign AI systems for critical national functions.